1
ISYE 6501 Final Exam Practice: 100 Questions with Answers
Section 1: Classification & Tree-Based Models (Questions 1–20)
Q1. What is the primary purpose of a CART (Classification and Regression Tree) model?
A. To cluster data into groups
B. To recursively partition data to predict a response
C. To smooth time series data
D. To optimize resource allocation
Answer: B
Rationale: CART recursively partitions data based on predictor variables to predict a
categorical or quantitative response. It is a nonparametric method.
Q2. In a classification tree, what is the typical rule of thumb for the minimum number of
data points in a leaf?
A. At least 1 point
B. At least 5% of the data points
C. At least 50 points
D. At least 10% of the data points
Answer: B
Rationale: A common rule of thumb is each leaf should have at least 5% of the data points to
avoid overfitting.
Q3. How is pruning typically performed in a CART model?
A. By removing branches that decrease variance in the training set
B. By using validation data to determine if branching improves estimation error
C. By increasing tree depth until each leaf has one point
D. By replacing quantitative variables with binary ones
Answer: B
Rationale: Pruning uses validation data to determine whether branching actually improves
estimation error, preventing overfitting.
Q4. What is the primary benefit of a Random Forest compared to a single Decision Tree?
A. Reduction in variance and improved accuracy through averaging
B. Faster computation time
C. Higher interpretability
D. Simpler implementation
Answer: A
Rationale: Random Forest reduces variance and improves accuracy by averaging multiple
bootstrapped decision trees.
, 2
Q5. In a soft SVM classifier, what is the tradeoff controlled by the parameter lambda?
A. Maximizing margin versus minimizing classification errors
B. Maximizing training accuracy versus test accuracy
C. Minimizing computation versus maximizing accuracy
D. Maximizing number of support vectors versus minimizing them
Answer: A
Rationale: The soft SVM classifier trades off between maximizing the margin and minimizing
classification errors. A small lambda emphasizes error minimization; a large lambda
emphasizes margin maximization.
Q6. What does a very small coefficient in an SVM model indicate about a predictor variable?
A. The variable is very important for classification
B. The variable is not very important for classification
C. The variable should be removed from the model
D. The variable causes overfitting
Answer: B
Rationale: A coefficient close to zero indicates that the predictor variable is not very
important to the classification.
Q7. Why is scaling or normalizing data critical for k-NN?
A. To reduce computational complexity
B. To prevent attributes with larger ranges from dominating distance calculations
C. To transform data into a normal distribution
D. To handle categorical variables
Answer: B
Rationale: Without scaling, attributes with larger numerical ranges dominate the distance
calculation in k-NN, biasing results.
Q8. What is the 2-norm distance?
A. Euclidean distance
B. Manhattan distance
C. The largest dimension value
D. The sum of absolute values
Answer: A
Rationale: The 2-norm is Euclidean (straight-line) distance.
Q9. What is the difference between standardization and scaling?
A. Scaling is bounded in range; standardization transforms to a normal distribution
B. They are the same thing
, 3
C. Standardization is bounded; scaling is not
D. Scaling removes outliers; standardization does not
Answer: A
Rationale: Scaling is bounded in range (e.g., 0 to 1), while standardization scales to a normal
distribution using (value − mean) / standard deviation.
Q10. What is overfitting?
A. When a model fits training data too closely and performs poorly on new data
B. When a model is too simple
C. When a model has high bias
D. When a model ignores training data
Answer: A
Rationale: Overfitting occurs when a model learns noise in training data, leading to poor
performance on unseen data.
Q11. What is the rule of thumb for minimum data points per leaf in a CART model?
A. At least 1 point
B. At least 5% of data points
C. At least 50 points
D. At least 10% of data points
Answer: B
Rationale: A common rule of thumb is each leaf should have at least 5% of the data points.
Q12. What is the rule of thumb for factor-based models regarding data points to factors?
A. At least 5 times as many data points as factors
B. At least 10 times as many data points as factors
C. At least 2 times as many data points as factors
D. Equal numbers
Answer: B
Rationale: A common rule of thumb is to have at least 10 times as many data points as
factors selected.
Q13. What does a confusion matrix show?
A. The performance of a classification model
B. The distribution of continuous variables
C. The correlation between features
D. The time series trend
Answer: A
Rationale: A confusion matrix shows the performance of a classification model.
ISYE 6501 Final Exam Practice: 100 Questions with Answers
Section 1: Classification & Tree-Based Models (Questions 1–20)
Q1. What is the primary purpose of a CART (Classification and Regression Tree) model?
A. To cluster data into groups
B. To recursively partition data to predict a response
C. To smooth time series data
D. To optimize resource allocation
Answer: B
Rationale: CART recursively partitions data based on predictor variables to predict a
categorical or quantitative response. It is a nonparametric method.
Q2. In a classification tree, what is the typical rule of thumb for the minimum number of
data points in a leaf?
A. At least 1 point
B. At least 5% of the data points
C. At least 50 points
D. At least 10% of the data points
Answer: B
Rationale: A common rule of thumb is each leaf should have at least 5% of the data points to
avoid overfitting.
Q3. How is pruning typically performed in a CART model?
A. By removing branches that decrease variance in the training set
B. By using validation data to determine if branching improves estimation error
C. By increasing tree depth until each leaf has one point
D. By replacing quantitative variables with binary ones
Answer: B
Rationale: Pruning uses validation data to determine whether branching actually improves
estimation error, preventing overfitting.
Q4. What is the primary benefit of a Random Forest compared to a single Decision Tree?
A. Reduction in variance and improved accuracy through averaging
B. Faster computation time
C. Higher interpretability
D. Simpler implementation
Answer: A
Rationale: Random Forest reduces variance and improves accuracy by averaging multiple
bootstrapped decision trees.
, 2
Q5. In a soft SVM classifier, what is the tradeoff controlled by the parameter lambda?
A. Maximizing margin versus minimizing classification errors
B. Maximizing training accuracy versus test accuracy
C. Minimizing computation versus maximizing accuracy
D. Maximizing number of support vectors versus minimizing them
Answer: A
Rationale: The soft SVM classifier trades off between maximizing the margin and minimizing
classification errors. A small lambda emphasizes error minimization; a large lambda
emphasizes margin maximization.
Q6. What does a very small coefficient in an SVM model indicate about a predictor variable?
A. The variable is very important for classification
B. The variable is not very important for classification
C. The variable should be removed from the model
D. The variable causes overfitting
Answer: B
Rationale: A coefficient close to zero indicates that the predictor variable is not very
important to the classification.
Q7. Why is scaling or normalizing data critical for k-NN?
A. To reduce computational complexity
B. To prevent attributes with larger ranges from dominating distance calculations
C. To transform data into a normal distribution
D. To handle categorical variables
Answer: B
Rationale: Without scaling, attributes with larger numerical ranges dominate the distance
calculation in k-NN, biasing results.
Q8. What is the 2-norm distance?
A. Euclidean distance
B. Manhattan distance
C. The largest dimension value
D. The sum of absolute values
Answer: A
Rationale: The 2-norm is Euclidean (straight-line) distance.
Q9. What is the difference between standardization and scaling?
A. Scaling is bounded in range; standardization transforms to a normal distribution
B. They are the same thing
, 3
C. Standardization is bounded; scaling is not
D. Scaling removes outliers; standardization does not
Answer: A
Rationale: Scaling is bounded in range (e.g., 0 to 1), while standardization scales to a normal
distribution using (value − mean) / standard deviation.
Q10. What is overfitting?
A. When a model fits training data too closely and performs poorly on new data
B. When a model is too simple
C. When a model has high bias
D. When a model ignores training data
Answer: A
Rationale: Overfitting occurs when a model learns noise in training data, leading to poor
performance on unseen data.
Q11. What is the rule of thumb for minimum data points per leaf in a CART model?
A. At least 1 point
B. At least 5% of data points
C. At least 50 points
D. At least 10% of data points
Answer: B
Rationale: A common rule of thumb is each leaf should have at least 5% of the data points.
Q12. What is the rule of thumb for factor-based models regarding data points to factors?
A. At least 5 times as many data points as factors
B. At least 10 times as many data points as factors
C. At least 2 times as many data points as factors
D. Equal numbers
Answer: B
Rationale: A common rule of thumb is to have at least 10 times as many data points as
factors selected.
Q13. What does a confusion matrix show?
A. The performance of a classification model
B. The distribution of continuous variables
C. The correlation between features
D. The time series trend
Answer: A
Rationale: A confusion matrix shows the performance of a classification model.
